warp.tile\_scatter\_add
=======================

.. function:: warp._src.lang.tile_scatter_add(a: Tile[Any, tuple[int, ...]], i: int32, value: Any, has_value: bool, atomic: bool) -> None

   .. hlist::
      :columns: 8

      * Kernel
      * Differentiable

   Scatter-add a per-thread value into a shared-memory tile.
   
   Cooperative operation -- all threads in the block must call this function.
   Each thread whose ``has_value`` is ``True`` adds ``value`` at index ``i``.
   
   A synchronization barrier is included so the updated values are visible to
   all threads after the call returns.
   
   :param a: A shared-memory tile to scatter-add into.
   :param i: Index of the element to add to.
   :param value: The value to add (must match the tile's dtype).
   :param has_value: Whether this thread should perform the add.
   :param atomic: If True (default), use atomic add for safe concurrent writes.
                  Set to False when indices are guaranteed unique across threads
                  (e.g., lane-parallel writes) for better performance.
   
   .. rubric:: Example
   
   .. code-block:: python
   
       @wp.kernel
       def histogram(data: wp.array[float], out: wp.array[float]):
   
           bins = wp.tile_zeros(dtype=float, shape=4, storage="shared")
           i = wp.tid()
           # Bin values in [0, 8) into 4 bins of width 2
           b = int(data[i] / 2.0)
           wp.tile_scatter_add(bins, b, 1.0, True)
           wp.tile_store(out, bins, offset=0)
   
       data = wp.array([0.5, 1.0, 2.5, 3.0, 4.5, 5.0, 6.5, 7.0], dtype=float)
       output = wp.zeros(4, dtype=float)
       wp.launch_tiled(histogram, dim=[1], inputs=[data, output], block_dim=8)
   
       print(output.numpy())
   
   .. code-block:: text
   
       [2. 2. 2. 2.]
   

.. function:: warp._src.lang.tile_scatter_add(a: Tile[Any, tuple[int, ...]], i: int32, j: int32, value: Any, has_value: bool, atomic: bool) -> None
   :noindex:

   .. hlist::
      :columns: 8

      * Kernel
      * Differentiable

   

.. function:: warp._src.lang.tile_scatter_add(a: Tile[Any, tuple[int, ...]], i: int32, j: int32, k: int32, value: Any, has_value: bool, atomic: bool) -> None
   :noindex:

   .. hlist::
      :columns: 8

      * Kernel
      * Differentiable

   

.. function:: warp._src.lang.tile_scatter_add(a: Tile[Any, tuple[int, ...]], i: int32, j: int32, k: int32, l: int32, value: Any, has_value: bool, atomic: bool) -> None
   :noindex:

   .. hlist::
      :columns: 8

      * Kernel
      * Differentiable